Self-Organizing Maps for identification of zeolitic diagenesis patterns in closed hydrologic systems: implications for Mars.

Abstract:

Zeolites, authigenic silicate minerals formed in alkaline environments, have been detected on Mars using orbital data, and these detections are mostly associated with the Fe, Mg-rich phyllosilicates, hydrated silica, or carbonates. Tuffs are a main precursor for zeolites in paleolake deposits on Earth, and the mineral assemblages found in these deposits are used to reconstruct the hydrochemical history of these environments.

We have studied mineral paragenesis in thirteen paleolake deposits from USA, Mexico, Greece, and Tanzania (either directly, or through literature review), and found that the mineral paragenesis in these sites is often complex and difficult to generalize due to the wide range of mineral assemblages even within a same site/same bed. We have thus collected X-ray Diffraction based bulk mineral assemblage determinations from 1648 tuff samples from previous and our studies. We applied the Kohonen Self-Organizing Maps (SOM) to look for interesting patterns in the dataset without prescribing any specific interpretation, and for information reduction and classification. We also used the Decision Tree (DT) method to characterize these clusters. We were able to define clear class boundaries between the fresh glass, non-analcime zeolites, analcime, and K feldspar. The non-analcime zeolites were grouped into several classes based on their specific zeolites. This study showed that smectite abundances increase relative to glass as alteration progresses. Smectite abundances are lower in samples with abundant analcime, and even lower in samples with authigenic K feldspar, likely reflecting higher degrees of saline-alkaline alteration. However, since smectites are identified in samples at all stages of alteration, their presence is not useful for distinguishing diagenetic facies. The study concludes that generalizing the complex geochemical behaviors using unsupervised statistical learning methods can help to identify the most prominent geochemical behaviors. The trained SOM will be used to characterize the places where zeolites and other authigenic minerals were identified using Mars orbital data. The model can be used as a data driven predictive model for future Mars missions studying possible paleolakes.